Flow scaffolding

Flow Scaffolding Templates

Standardized templates for creating new Claude Code plugin components. Each template includes required frontmatter, section structure, and placeholder content that follows project conventions.

When to Apply

Reference these templates when:

• Creating new agents, commands, skills, or hooks via /create-flow
• Manually scaffolding components and need the correct structure
• Validating existing components against standard structure

Component Templates

1. Command Template

File: templates/command.md

Creates a command in commands/{name}.md with:

• YAML frontmatter (name, description, category, complexity, mcp-servers, personas)
• Standard sections: Triggers, Usage, Behavioral Flow, Tool Coordination, Examples, Boundaries
• Optional AI collaboration section

2. Agent Template

File: templates/agent.md

Creates an agent in agents/{name}.md with:

• YAML frontmatter (name, description, tools)
• Standard sections: Role, Analysis Workflow, Decision Matrices, Output Format, Boundaries
• Step-by-step workflow with Glob/Grep patterns

3. Skill Template

File: templates/skill.md

Creates a skill directory skills/{name}/ with:

• SKILL.md with frontmatter (name, description)
• Standard sections: When to Apply, Domain Content, How to Use, References
• Optional references/ and templates/ directories

4. Hook Template

File: templates/hook.json

Creates a hook entry for claude/settings.json with:

• Matcher expression for tool/condition filtering
• Hook command (Node.js script)
• Description

5. Plugin Template

File: templates/plugin.json

Creates a plugin manifest in .claude-plugin/plugin.json with:

• Plugin metadata (name, description, version, author)
• Directory structure for bundled skills, agents, hooks, MCP servers

Template Complexity Levels

Level	Description	Sections	Use Case
minimal	Core structure only	Required sections with minimal content	Quick prototyping
standard	Complete structure	All required + common optional sections	Most components
advanced	Full-featured	All sections + AI collaboration + references + plugin manifest	Complex workflows

Naming Conventions

Component	File Pattern	Name Format
Command	commands/{name}.md	kebab-case verb-noun (e.g., deploy-check)
Agent	agents/{name}.md	kebab-case role-noun (e.g., security-scanner)
Skill	skills/{name}/SKILL.md	kebab-case domain-noun (e.g., redis-best-practices)
Hook	Entry in settings.json	descriptive sentence
Plugin	.claude-plugin/plugin.json	kebab-case (e.g., my-plugin)

Frontmatter Field Reference

Command Fields (Legacy)

As of Claude Code v2.1.3, commands and skills are unified. New components should use skill format. Command format is maintained for backward compatibility.

Field	Required	Type	Example
name	Yes	string	"deploy-check"
description	Yes	string	"Pre-deployment validation..."
category	Yes	string	"utility" / "review" / "workflow"
complexity	Yes	string	"basic" / "advanced"
mcp-servers	Yes	array	[] or ["grafana"]
personas	Yes	array	[] or ["arch-reviewer"]

Skill Fields

Field	Required	Type	Example
name	No*	string	"redis-best-practices"
description	Recommended	string (multiline)	"Redis optimization guide..."
argument-hint	No	string	"[issue-number]"
disable-model-invocation	No	boolean	true
user-invocable	No	boolean	false
allowed-tools	No	array	["Read", "Grep"]
model	No	string	"sonnet" / "opus"
context	No	string	"fork"
agent	No	string	"Explore" / "Plan"
hooks	No	object	{PreToolUse: [...]}

*Defaults to directory name if omitted.

Agent Skills Open Standard Fields (Optional)

Field	Type	Example
license	string	"MIT"
compatibility	array	["claude-code", "cursor", "gemini-cli"]
metadata	object	{version: "1.0", category: "review"}

Agent Fields

Field	Required	Type	Example
name	Yes	string	"security-scanner"
description	Yes	string	"Security analysis specialist..."
tools	No	array	["Read", "Grep", "Glob", "Bash"]
disallowedTools	No	array	["Write", "Edit"]
model	No	string	"sonnet" / "opus" / "haiku" / "inherit"
permissionMode	No	string	"acceptEdits" / "plan"
maxTurns	No	number	50
skills	No	array	["my-skill"]
mcpServers	No	array	["server-name"] or inline config
hooks	No	object	{PreToolUse: [...]}
memory	No	string	"user" / "project" / "local"
background	No	boolean	true
isolation	No	string	"worktree"

Plugin Fields (plugin.json)

Field	Required	Type	Example
name	Yes	string	"my-plugin"
description	Yes	string	"Plugin description..."
version	Yes	string	"1.0.0"
author	No	object	{"name": "Author Name"}
homepage	No	string	URL
repository	No	string	Git URL
license	No	string	"MIT"

String Substitution Variables

Available in skill body content:

Variable	Description
$ARGUMENTS	All arguments passed to the skill
$ARGUMENTS[N] / $N	Specific argument by 0-based index
${CLAUDE_SESSION_ID}	Current session ID
${CLAUDE_SKILL_DIR}	Directory containing the skill's SKILL.md

Dynamic context injection via shell:

!`command`    # Output replaces placeholder before Claude sees content

Skill Loading Priority

Priority	Location	Scope
1	Enterprise (managed settings)	Organization-wide
2	Personal (~/.claude/skills/)	All projects
3	Project (.claude/skills/)	Current project
4	Plugin (<plugin>/skills/)	Namespaced

Section Content Guidelines

Triggers

• Use concrete action phrases, not vague descriptions
• List 4-6 specific scenarios
• Each trigger should be independently sufficient to activate the component

Behavioral Flow

• Number each phase sequentially
• Use bold phase names: Phase 1: Discovery
• Include sub-steps with tool-specific instructions
• Decision matrices for branching logic

Examples

• Minimum 2 examples per command
• Include both simple and complex usage
• Show expected output or behavior in comments
• Cover edge cases in advanced examples

Boundaries

• Will: 5-8 specific capabilities
• Will Not: 3-5 explicit limitations
• Use active voice, be specific about what is in/out of scope